Partners who understand AIGC and GPT models should know more or less about the situation: in the field of AIGC at home and abroad, all large and small manufacturers and AI entrepreneurs are in full swing to carry out the "arms race". And just last two weeks, the Baidu Wenxin internal test account that I applied for for more than a month was finally issued. So, what is the AI drawing level of the multimodal GPT model released by Baidu some time ago - Wen Xin Yiyan? Let's try it together~
To facilitate the quantification of drawing quality, I will score from 3 angles of [Answer Fit], [Drawing Quality], and [Creative Innovation] (all out of 10 points), of course, this is with personal subjective opinion~
1 Characters
CASE1 Little girl selling matches
[Answer Fit] 10, [Drawing Quality] 8, [Creative Innovation] 9
Don't say, this is quite like that, if the copyright problem is solved, can this picture be directly placed in the book for illustration, and the wall as a poster? One more and see how it performs?
[Answering Fit] 10, [Drawing Quality] 10, [Creative Innovation] 10
It feels even better...
CASE2 Handsome Guy Who Reads (Anime)
[Answer Fit] 9, [Drawing Quality] 8, [Creative Innovation] 8
It still meets the appeal of text description.
CASE3 Beautiful girl playing the piano
And let it draw "beautiful girls playing the piano"?
[Answering fit] 10, [drawing quality] 9, [creative innovation] 9
Here's another one and see how?
[Answering fit] 10, [drawing quality] 6, [creative innovation] 9
I feel that the eyes are a little unnatural, everything else is fine~
Let's do this for the character chapter~
2 Natural landscapes
CASE4 Landscape with nine curves and eighteen bends
[Answering fit] 10, [drawing quality] 9, [creative innovation] 9
There are curved rivers, mountains, and water, and the title fits perfectly. Overall, the title conformance and artistic conception are quite good. Afraid of its pride, a painting with a full score of 10, give a 9 points.
CASE5 Seeing the clouds
[Answering fit] 8, [drawing quality] 9, [creative innovation] 9
3 Idioms and Imagination
CASE6 Grasshopper Shakes Tree
[Answer Fit] 8, [Drawing Quality] 7, [Creative Innovation] 7
To be honest, there is a loss of ugliness...
CASE7 Elephants bloom
[Answering fit] 8, [drawing quality] 9, [creative innovation] 9
No problem, this is indeed an elephant blossom
CASE8 Pig pig cabbage
[Answering fit] 10, [drawing quality] 9, [creative innovation] 9
CASE9 Mandarin duck hot pot
As a Sichuanese, how about taking a look at the "Mandarin Duck Hot Pot" that you often order? Ha ha
[Answering fit] 10, [drawing quality] 9, [creative innovation] 9
No problem, old iron! This is literally "Mandarin duck" + "hot pot" ~
CASE10 Egg Scones
[Answering fit] 10, [drawing quality] 9, [creative innovation] 9
CASE11 Fat boy with a tiger head
CASE12 Data Ocean
[Answer Fit] 8, [Drawing Quality] 8, [Creative Innovation] 8
CASE13 Data Companion
How about thinking of a logo for me? "DataSoulmate" ~
[Answering fit] 10, [drawing quality] 9, [creative innovation] 9
Nice, it's a cyberpunk blast!
brief summary
1. Objective evaluation of Baidu Wenxin's AI drawing ability. Actually, I also asked some other questions, but the generated ones were a bit outrageous. This failure rate is roughly around 30%. However, compared with the traditional, human drawings and design drawings, the efficiency of AI drawing quickly according to human needs is difficult for the former.
2. Tools and models for AIGC to generate diagrams. In addition to Baidu's Wen Xin, the large models of AIGC drawings are more well-known:
- midjourney - https://www.midjourney.com/
- Stable Diffusion - Stable Diffusion Online
- Adobe Firefly - https://www.adobe.com/tw/sensei/generative-ai/firefly.html
- ...
3. Application scenarios of AI drawing. The scenario that comes to mind for the time being:
- New media, illustration and illustration of writing
- Publishing field, illustration/illustration
- In the field of short video, quickly generate the cover or keyframe of the core idea
- In the field of advertising and marketing, creative ideas are issued and creative posters are generated
4. Risks or problems to be solved in the field of AI mapping:
- Data rights confirmation & intellectual property/copyright attribution issues
- The fit of the topic needs to be improved (there are still many technical challenges in this area)
- Difficulty 1: Fully understand user needs
- Difficulty 2: Completely translate the demand into reasonable, high-quality image content
- For biometrics (face recognition、... ) poses a risk of online fraud.
- For designers (graphic designers, UI and UE designers、... professional challenges
Thank you for reading, I am a software engineer who is committed to contributing to the improvement of social productivity by exploring, researching, sharing software tools, or engaging in software development~
If you also like this article, you can also like, collect, comment, and retweet ~I wish you a happy life~